Cursor 2.1 Agent can't access in-IDE browser

[NullPoint]

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

After confirming that the browser is enabled under Tools → Browser Automation, when asking Agent to use the browser it claims it does not have access to that tool and tries to use one of the other MCP browsers I have setup. Eventually I figured out that clicking the world icon in the agent panel to open a browser tab is not enough. Until you type in the @browser command the agent does not have access to the tools.

After typing the @browser command the agent says it can see the tools now but they return a connection error. I’m able to load pages in the in IDE browser tab so it’s not network.

I tried enabling/disabling other MCP browser tools and eventually discovered, unless it’s coincidence, that enabling Chrome DevTools MCP seems to help. After enabling it the Agent can now at least see that I have an IDE browser tab open and it can take screenshots but it cannot navigate or take snapshots so it’s still essentially useless.

Steps to Reproduce

The above happens consistently for me but I assume there would be posts about it if it was happening for everyone. Might be difficult to reproduce.

Expected Behavior

I believe it’s meant to work as described here:

Browser | Cursor Docs

Operating System

Windows 10/11

Current Cursor Version (Menu → About Cursor → Copy)

Version: 2.1.26 (user setup)
VSCode Version: 1.105.1
Commit: f628a4761be40b8869ca61a6189cafd14756dff0
Date: 2025-11-24T05:39:06.655Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Windows_NT x64 10.0.26200

For AI issues: which model did you use?

Auto

Additional Information

Here is the agent’s take on what works and what doesn’t:

Working tools:
browser_take_screenshot — can capture screenshots
browser_console_messages — can read console messages
browser_network_requests — can see network requests
browser_wait_for — can wait for time/conditions
browser_resize — can resize the browser window
Not working:
browser_navigate — can’t navigate to URLs
browser_snapshot — can’t get page structure for interaction
browser_click, browser_type, browser_hover, browser_select_option — require snapshot, so they won’t work
browser_press_key — doesn’t work
browser_navigate_back — doesn’t work

However I didn’t see evidence of a resize when it ran that command.

Does this stop you from using Cursor

No - Cursor works, but with this issue

[sanjeed5 (Sanjeed)]

Thanks for the detailed report - this partial tool availability is unusual. The read-only tools working but navigation/interaction tools failing suggests an issue with the browser automation server connection.

Could you share:

1. Request IDs: CMD + Shift + P > Report AI Action > select a failed browser_navigate or browser_snapshot action
2. Console errors: Help > Toggle Developer Tools > Console tab when you get the connection error
3. Tools status: What does Settings > Tools & MCP > Browser Automation say? (“Ready - Chrome detected” or something else?)
4. Both browser options: Does this happen with both “Browser Tab” and “Google Chrome” in the agent dropdown?

[NullPoint]

The in-IDE browser started (mostly) working after a few nightly updates. Most actions work anyway… resizing never works, which probably has to do with the browser living in an IDE tab. The agent, however, thinks that resizing works as none of the tools are reporting errors now. That should be straightforward to reproduce, probably a known issue.

Currently the agent gets confused at times when using the tools, for example believing a click happened when it didn’t. That could be because the current model being served up by auto is clearly ancient by LLM standards.

You guys know that going from a frontier model to a model multiple generations old from one day to the next is bad PR right?